TypeError: chatMessage._getType is not a function

[riccardolinares]

I m following the YT tutorial, but after the first message sent correctly, the second one give me this error:

error TypeError: chatMessage._getType is not a function
    at file:///Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/langchain/dist/chains/conversational_retrieval_chain.js:67:33
    at Array.map (<anonymous>)
    at ConversationalRetrievalQAChain.getChatHistoryString (file:///Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/langchain/dist/chains/conversational_retrieval_chain.js:66:18)
    at ConversationalRetrievalQAChain._call (file:///Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/langchain/dist/chains/conversational_retrieval_chain.js:90:60)
    at ConversationalRetrievalQAChain.call (file:///Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/langchain/dist/chains/base.js:65:39)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async handler (webpack-internal:///(api)/./pages/api/chat.ts:45:26)
    at async Object.apiResolver (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/api-utils/node.js:372:9)
    at async DevServer.runApi (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/next-server.js:513:9)
    at async Object.fn (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/next-server.js:815:35)
    at async Router.execute (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/router.js:243:32)
    at async DevServer.runImpl (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/base-server.js:432:29)
    at async DevServer.run (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/dev/next-dev-server.js:814:20)
    at async DevServer.handleRequestImpl (/Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/base-server.js:375:20)
    at async /Users/riccardolinares/Projects/gpt4-pdf-chatbot-langchain/node_modules/next/dist/server/base-server.js:157:99

Version:
"langchain": "^0.0.91",

[borel]

Hey @riccardolinares , how did you manage chat history ?

maybe you mixmatch custom and default chat memory history.

[riccardolinares]

I am working with Next.js and I create an API Route that take the question and the history from the body of the request.

api/chat:

  const { question, history } = req.body;

    const response = await chain.call({
      question: question,
      chat_history: history || [],
    });

On the page this is my code:

index.tsx

const [messageState, setMessageState] = useState<{
   messages: Message[];
   pending?: string;
   history: [string, string][];
   pendingSourceDocs?: Document[];
 }>({
   messages: [
     {
       message: 'Hi, what would you like to learn about this document?',
       type: 'apiMessage',
     },
   ],
   history: [],
 });

 const { messages, history } = messageState;

and if the response from the API Route is ok, I update the MessageState as follow:

try {
     const response = await fetch('/api/chat', {
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
       },
       body: JSON.stringify({
         question,
         history,
       }),
     });
     const data = await response.json();
     console.log('data', data);

     if (data.error) {
       setError(data.error);
     } else {
       setMessageState((state) => ({
         ...state,
         messages: [
           ...state.messages,
           {
             type: 'apiMessage',
             message: data.text,
             sourceDocs: data.sourceDocuments,
           },
         ],
         history: [...state.history, [question, data.text]],
       }));
     }

I am not sure if it is clear enough... in case let me know

[borel]

Hey @riccardolinares , what kind of chain are you using

    const response = await chain.call({
      question: question,
      chat_history: history || [],
    });

Not sure it expect to pass directly the chat_history like that

[riccardolinares]

These are my chain and model:

const model = new OpenAI({
    temperature: 0, // increase temepreature to get more creative answers
    modelName: 'gpt-3.5-turbo', //change this to gpt-4 if you have access
  });

  const chain = ConversationalRetrievalQAChain.fromLLM(
    model,
    vectorstore.asRetriever(),
    {
      qaTemplate: QA_PROMPT,
      questionGeneratorTemplate: CONDENSE_PROMPT,
      returnSourceDocuments: true, //The number of source documents returned is 4 by default
    },
  );

btw it seems like the _getType() function is not found at all :/

[bryceamacker]

Also getting this issue, I believe it's a new breaking change but not sure where it's introduced.

[borel]

Hi @bryceamacker @riccardolinares

The documentation is pretty bad for this case .

There is two solutions to manage custom history using ConversationalRetrievalQAChain.

1. First you can use a chat_history as a string and use concatenation like this
   const chatHistory = question + res.text;

2. You can use an array as chat_history but you have to use the required langchain model ( https://js.langchain.com/docs/modules/schema/chat-messages ) like this
   const chatHistory = [new HumanChatMessage("Is Karim Benzema is a great plyer"),new AIChatMessage("Yes its is !")];

[siddarthvader]

this works, thanks for the help @borel

[riccardolinares]

I was using an array of strings to manage the chat history. Converting the array into a single string solved my problem! :)

Thanks!

[davideuler]

I came across this issue too after upgrade langchain to 0.0.90. And I fixed the issue in my fork.

https://github.com/davideuler/gpt4-pdf-chatbot-langchain-chromadb

[esponges]

I came across this issue too after upgrade langchain to 0.0.90. And I fixed the issue in my fork.

https://github.com/davideuler/gpt4-pdf-chatbot-langchain-chromadb

This is a clean approach, using @borel second strategy. Thanks!

[kirandash]

This worked for me:

import { AIMessage, HumanMessage } from "@langchain/core/messages";
chatHistory.push(new HumanMessage(initialPrompt), new AIMessage(response.text));